Simulating a single chatbot is easy. Simulating a society is not.
A new startup founded by Stanford researchers wants to model entire populations using AI agents trained on real human data — and investors just handed it $100 million to try.
Simile AI emerged from stealth Thursday with a $100 million Series A led by Index Ventures, positioning itself as infrastructure for what it calls “society-scale simulation.”
What Just Happened
Simile was founded by Stanford researchers behind the 2023 “Smallville” project, an academic experiment that used large language model–powered agents to simulate a small virtual town. That research drew attention for showing how generative AI agents could form relationships, organize events, and exhibit emergent social behaviors.
Now, Simile is commercializing the concept.
The company says its platform trains AI agents on structured human data — including anonymized interview data from Gallup — to create digital twins of individuals and demographic groups. These agents can then be deployed in simulated environments to test how people might respond to decisions, policies, or product changes.
Early customers include Wealthfront and CVS Health, which are using the system to model customer reactions and strategic decisions in a low-risk environment.
Among its backers is AI researcher Andrej Karpathy, who publicly congratulated the company on X and described the approach as exploring an “under-explored dimension” of large language models — moving beyond single, static chatbot personalities toward population-level simulations.
Simile says its long-term goal is to model trillions of interactions across societies.
Why This Matters for the US
For US companies, this is more than an academic exercise.
If Simile’s approach works, it could fundamentally change how American businesses test products, roll out pricing strategies, or assess regulatory impact. Instead of relying solely on focus groups, surveys, or A/B tests, companies could simulate thousands — or millions — of AI agents representing real demographic slices of the US population.
That has obvious implications for sectors like finance, healthcare, retail, insurance, and public policy — all industries where misreading public reaction can cost billions.
For startups, it lowers the cost of experimentation. For enterprise companies, it offers a potential risk buffer before launching controversial decisions. For policymakers, it introduces the possibility of testing policy scenarios before implementation — though that raises serious governance and bias questions.
In a US market where AI spending is projected to surpass $300 billion annually by the end of the decade, tools that promise predictive behavioral modeling could quickly become core enterprise infrastructure.
Expert Analysis: The Next Layer of AI Infrastructure?
Most AI investment over the past three years has focused on model training, inference efficiency, copilots, and agentic workflows.
Simile is attacking a different layer: population simulation.
The original Smallville experiment demonstrated that LLM-based agents can exhibit believable social dynamics when given memory, goals, and environmental context. What Simile appears to be doing is combining that agent architecture with structured human data to anchor simulations in real-world demographics.
That’s a powerful — and potentially controversial — leap.
Training agents on interview datasets like Gallup’s could help anchor behavior patterns statistically. But it also raises questions around representativeness, bias amplification, and how accurately simulated agents reflect marginalized communities in the US.
If simulations influence business or policy decisions, the governance framework around them will matter as much as the tech itself.
From Digital Twins to Digital Societies
The idea of “digital twins” isn’t new. Manufacturing, aerospace, and logistics companies have long built virtual replicas of machines or supply chains to test outcomes.
Simile extends that concept to human systems.
While major AI labs such as OpenAI and Google DeepMind focus on scaling model intelligence, Simile is focused on scaling social interaction layers on top of those models.
It’s a subtle but meaningful shift: from building smarter individual models to modeling the emergent behavior of many interacting agents.
That could become increasingly relevant as AI systems begin making autonomous decisions in financial markets, healthcare triage, or urban planning simulations.
What Happens Next
With $100 million in fresh capital, Simile will likely invest heavily in compute infrastructure, dataset expansion, and enterprise partnerships.
The near-term test will be whether companies like Wealthfront and CVS Health can demonstrate measurable ROI from simulation-driven decision-making.
Longer term, regulators may start paying attention. If companies rely on AI-simulated populations to make consequential decisions, transparency around methodology and dataset composition could become a policy issue — particularly in Washington, D.C., where AI governance frameworks are still evolving.
Final Take
Simile is betting that the next frontier of AI isn’t just smarter chatbots — it’s synthetic societies.
The $100 million raise suggests investors believe behavioral simulation could become foundational infrastructure in an AI-driven economy.
For the US tech ecosystem, the question isn’t whether population-scale modeling will happen. It’s who will control it, how it’s governed, and whether simulated citizens will meaningfully reflect the real ones businesses and policymakers serve.